Automatic Recognition of Impact Craters on the Martian Surface from DEM and Images - Planetary Exploration and Science: Recent Results and Advances - page 112

Tabl e 6. 2 Detection rates

Region GT TD TDR FD FDR

<5km 344 256 74.4 125 32.8

5-10 km 190 123 64.7 72 36.9

10-15 km 65 52 80 15 22.4

15-20 km 26 21 80.8 3 12.5

20-25 km 15 13 86.6 4 23.5

>25 km 32 30 93.8 4 11.8

To t a l 672 495 73.4 223 31.6

Tabl e 6. 3 Detection rates

Region TDR FDR

A 55.2 7.8

B 73.3 6.8

data. In this work by Bue and Stepinski ( 2007 ), the test site is chosen in a
heavily cratered terrain covering almost 1.0 10 6 km 2 . The craters listed in Barlow
catalogue serve as the ground truth to evaluate the performance of this method.
A method based on the geomorphological features of Martian craters is intro-
duced by Xie et al. ( 2013 ). In Table 6.3 , we list the performance of this method.

6.4.2 Discussion

There are two categories of crater-detection algorithm, which are image based and
topography based separately. We have presented these methods and shown their
performances. Generally speaking, better detection rate can be obtained when the
approach is based on images rather than topography. It is because the resolution
of planetary images is always higher than topography contributing to detecting
more small craters. The heavily degraded craters and overlapped craters are difficult
to be detected in DEM because of the limited resolutions. As more and more
collections of images and topography data at higher spatial resolution can be
obtained, more efficient detection methods are very necessary, even combining
images with topography data.

6.5 Conclusion

As some algorithms have shown their good performances, the successful detection
of fresh well-formed craters is not hard to achieve. However, the detection of eroded
craters as well as the craters which are partially erased is challenging because,
with the increase of sensitivity, the number of false detections increases as well. In